Radioactive aerosol real-time monitoring method and system

Abstract

The invention relates to the technical field, and discloses a radioactive aerosol real-time monitoring method and system, and the method comprises the steps: collecting real-time multi-modal data of aerosol monitoring in a target region, and obtaining the fusion features of the real-time multi-modal data; performing nuclide identification on the fusion features of the real-time multi-modal data by using a pre-trained identification model to obtain an identification result; on the basis of the recognition result, semantic attributes of nuclides of known types are obtained by judging the semantic attributes of the nuclides, and semantic attributes of nuclides of unknown types are speculated by using a zero sample learning model; and based on the attribute of the nuclide corresponding to the real-time multi-modal data, generating a suggested decision by using a predefined decision support model. Through real-time monitoring and rapid identification, risk early warning can be given out in time, a scientific basis is provided for emergency response, and the problem that sudden radiation events cannot be handled in time is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental monitoring, and in particular to a real-time monitoring method and system for radioactive aerosols. Background Art

[0002] With the continuous advancement of science and technology and the enhancement of environmental protection awareness, the monitoring and management of radioactive aerosols are particularly important. Radioactive aerosols refer to tiny particles of radioactive substances suspended in the air. They can be spread to far places through the atmosphere, posing a potential threat to human health and the ecological environment. Therefore, real-time monitoring of the types, concentrations, and distribution of radioactive aerosols is of great significance for taking timely response measures and reducing the risk of radioactive contamination.

[0003] Traditional radioactive aerosol monitoring methods mainly rely on a single monitoring method, such as collecting aerosol samples through air samplers and then using radioactive analysis instruments for offline analysis. This method has the disadvantages of long monitoring cycle, poor real-time performance, and complex operation, which makes it difficult to meet the real-time and accuracy requirements of modern environmental monitoring.

[0004] Therefore, it is necessary to provide a real-time monitoring method and system for radioactive aerosols to solve the above technical problems. Summary of the invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a real-time monitoring method for radioactive aerosols, which combines multimodal data fusion technology, advanced machine learning algorithms and decision support systems to achieve comprehensive, real-time and accurate monitoring and evaluation of radioactive aerosols.

[0007] To solve the above technical problems, the present invention provides the following technical solutions: a method for real-time monitoring of radioactive aerosols, comprising: collecting real-time multimodal data of aerosol monitoring in a target area, and obtaining fusion features of the real-time multimodal data; using a pre-trained recognition model to perform nuclide identification on the fusion features of the real-time multimodal data to obtain a recognition result; based on the recognition result, by judging the semantic attributes of the nuclide, obtaining the semantic attributes of the known type of nuclide, and using a zero-sample learning model to infer the semantic attributes of the unknown type of nuclide; based on the attributes of the nuclide corresponding to the real-time multimodal data, using a predefined decision support model to generate a recommended decision.

[0008] As a preferred solution of the real-time monitoring method of radioactive aerosols described in the present invention, the real-time multimodal data includes: through the sensor modules deployed in the target area, the multimodal data of radiation data, meteorological data, air quality data and location data are collected in real time, and the feature extraction method is used to process each modal data separately. After the features are extracted, all single modal features are integrated to form a highly generalized fusion feature.

[0009] As a preferred solution of the real-time monitoring method of radioactive aerosols described in the present invention, the method of obtaining fusion features of real-time multi-modal data includes applying a specific feature extraction method to each modal data, specifically:

[0010] The radiation data uses spectrum analysis and statistical methods to extract peak energy and energy spectrum distribution characteristics;

[0011] Meteorological data by calculating trends, cyclical changes and outliers;

[0012] Air quality data identifies concentration fluctuation patterns by analyzing the time series characteristics of pollutants;

[0013] The location data were analyzed using geographic information system by calculating the distance matrix and spatial clustering between each monitoring point;

[0014] A multimodal fusion network is used to fuse the feature information of each modality, integrate the feature information of each modality, and form a fusion feature that comprehensively reflects the environmental status of the target area. Through the multimodal fusion network, the feature vectors of different modalities are used as input. MMFN contains multiple branches, each branch processes a corresponding modality feature, extracts high-level abstract features, and the outputs of each branch are converged at the fusion layer;

[0015] In the fusion layer, the outputs of each branch are aggregated and a unified fusion feature vector is generated through weighted summation. Finally, the fusion feature vector undergoes further nonlinear transformation and dimensionality reduction to form a highly concentrated and easy-to-interpret fusion feature.

[0016] As a preferred solution of the method for real-time monitoring of radioactive aerosols described in the present invention, the pre-trained recognition model includes: analyzing the fusion features through the pre-trained recognition model, outputting the predicted probability of each nuclide type, that is, the confidence, to provide a basis for subsequent nuclide type judgment, and sending the obtained fusion features as input to the trained neural network model, the neural network model is trained using a back propagation algorithm, and the prediction error is minimized by optimizing the loss function to learn the characterization patterns of different nuclide types;

[0017] The forward propagation process includes the input layer receiving the fused feature vector H and introducing the attention mechanism layer to dynamically weight the features of each modality:

[0018] A=Attention(H)

[0019] Among them, A represents the output of the attention mechanism layer, and Attention is the attention function;

[0020] Assume that there are L hidden layers, each with n l Each layer performs feature transformation and nonlinear activation functions to gradually extract higher-level features:

[0021] a (l) =Re(W (l) a (l-1) +b (l) +Dropout(a (l-1) , p l )+A (l-1) )

[0022] Among them, a (l) represents the activation output of the lth layer, Re represents the nonlinear activation function, W (l) represents the weight matrix connecting the l-1th layer and the lth layer, A (l-1) represents the output of the l-1 attention mechanism layer, p l Indicates the Dropout ratio of the lth layer, Dropout(a (l-1) , p l ) represents the random dropout function, b (l) Represents the bias vector of the lth layer; during the training process, the Dropout ratio p l Gradually decrease in each iteration to achieve dynamic adjustment:

[0023] p l =p start -iteration×decay

[0024] Among them, p start Represents the initial Dropout ratio, iteration represents the current number of iterations, and decay represents the decay rate of the Dropout ratio;

[0025] The output layer uses the Softmax function to output the predicted probability of each nuclide type, that is, the confidence

[0026]

[0027] Among them, Softmax means that the output is a function of probability distribution, W (L+1)represents the weight matrix connecting the last hidden layer and the output layer, a (L) represents the activation output of the output layer, b (L+1) represents the bias vector of the output layer, A (L) Represents the output of the current attention mechanism layer;

[0028] The training dataset contains a large amount of labeled multimodal data, each of which is annotated with the actual nuclide type, for model training and verification.

[0029] As a preferred embodiment of the method for real-time monitoring of radioactive aerosols described in the present invention, the method further comprises: the method for performing nuclide identification comprises screening out the confidence exceeding a preset confidence threshold from the confidence of each nuclide type corresponding to the real-time multimodal data, finding the nuclide type with the highest confidence in the prediction results after screening, and treating the nuclide type as the nuclide type identified by this monitoring, judging whether the nuclide type of the real-time multimodal data is known based on the known nuclide list, and obtaining nuclides of known types and nuclides of unknown types;

[0030] The initial confidence threshold is 0.9. Within 5 consecutive iterations, when the number of nuclide types exceeding the confidence threshold exceeds 50%, the confidence threshold increases by 0.05; when it is less than 30%, the confidence threshold decreases by 0.05, but the minimum value is 0.7. After each iteration, the confidence threshold is automatically adjusted according to the above rules.

[0031] The first-level nuclide type confidence threshold is 0.95, the second-level nuclide type confidence threshold is 0.9, the third-level nuclide type confidence threshold is 0.85, and the fourth-level nuclide type confidence threshold is 0.8. Classification is performed according to the nuclide type confidence corresponding to the real-time multimodal data. In the time series of multimodal data, the time series threshold is 3 iteration cycles. Within 3 consecutive iteration cycles, when the confidence of a certain nuclide type continues to increase and exceeds the time series threshold, this nuclide type is given priority.

[0032] Directly obtain the semantic properties of known nuclides or infer the properties of unknown nuclides using zero-shot learning models

[0033] Get a list of known nuclides, which contains all known nuclide types and their related attributes. Compare the determined nuclide type with the known nuclide list. If a match is successful, it is identified as a nuclide of known type; if no match is found, it is considered to be an unknown type of nuclide. Through the known / unknown classification of nuclide types, different processing strategies are adopted according to different types of nuclides.

[0034] As a preferred solution of the real-time monitoring method of radioactive aerosols described in the present invention, wherein: the determining the semantic attribute of the nuclide includes integrating the fusion features corresponding to the sample data and the nuclide of unknown type, and inputting the pre-trained zero-sample learning model;

[0035] Extract key attributes from the collected sample data to form an attribute matrix, merge the feature vector of the fusion feature corresponding to the unknown type of nuclide with the attribute matrix to form a unified input vector.

[0036] The input vector is sent to the pre-trained zero-shot learning model. The model infers the semantic attributes of the unknown type of nuclides by learning the attribute associations and patterns between different nuclides. The original zero-shot learning model uses the cross-nuclide attribute associations and patterns learned in the training phase to predict the semantic attributes of the unknown type of nuclides. The inference formula is:

[0037] f(X,θ)=S

[0038] Among them, f is the attribute prediction function, θ is the parameter of the zero-shot learning model, S is the semantic attribute of the predicted unknown type of nuclide, and X is the fused feature vector of the unknown type of nuclide; the original zero-shot learning model is optimized to obtain the improved model;

[0039] Since the value range of the output S is [0, 1], 0 indicates that it does not have the current semantic attribute, and 1 indicates that it has the current semantic attribute, the possibility of predicting the semantic attribute of the unknown nuclide increases with the increase of the output value.

[0040] As a preferred solution of the real-time monitoring method of radioactive aerosols described in the present invention, wherein: the generating of the recommended decision includes integrating the basic attributes and extended attributes of the nuclides obtained based on the real-time multimodal data into a comprehensive attribute list; collecting the basic attributes of the nuclides parsed from the real-time multimodal data, integrating the extended attributes, and forming a comprehensive attribute list according to the attribute information in a predefined format and structure;

[0041] Based on the generated comprehensive attribute list, the risk index is calculated to reflect the potential danger level of the nuclide. According to the preset risk level classification standard, the calculated risk index is mapped to a specific risk level, including low risk, medium risk, and high risk;

[0042] RiskIndex(N)=α1·R+α3·T+α4·T x

[0043] Among them, RiskIndex(N) represents the toxicity of the nuclide, α1, α2, α3, and α4 represent the weight coefficients in the risk index calculation, R represents the radioactivity intensity of the nuclide, T represents the half-life of the nuclide, and T xIndicates the toxicity of the nuclide;

[0044] Set the threshold of level 1 risk as RiskIndex l , the threshold of the secondary risk is RiskIndex m , the threshold of the third-level risk is RiskIndex h , when RiskIndex(N)<RiskIndex l When RiskIndex l ≤RiskIndex(N)<RiskIndex m When RiskIndex(N)≥RiskIndex h When the nuclides are classified into three levels of risk, the secondary and tertiary nuclide types will compete with each other in the three-level classification. Through further analysis and comparison, fuzzy logic and similarity evaluation are used to distinguish the nuclide types.

[0045] Establish a policy library containing response strategies. Each strategy corresponds to different risk levels and scenarios. When the risk level is determined, query the policy library for the matching response strategy. According to the actual situation and resource availability, formulate an execution plan for the response strategy, including the execution sequence, participants, and required materials. Execute the selected response strategy, monitor the execution effect, and adjust the strategy or take additional measures when necessary.

[0046] The risk levels and selected response strategies are collated into a report and communicated to operations personnel.

[0047] Another object of the present invention is to provide a radioactive aerosol real-time monitoring system, as a preferred solution of the radioactive aerosol real-time monitoring system described in the present invention, which includes: a fusion feature acquisition module, a nuclide identification module, a nuclide attribute acquisition module, and a decision generation module;

[0048] The fusion feature acquisition module is used to collect real-time multimodal data of aerosol monitoring in the target area and obtain fusion features of the real-time multimodal data;

[0049] The nuclide identification module is used to perform nuclide identification on the fusion features of the real-time multimodal data using a pre-trained identification model to obtain an identification result, wherein the identification result includes nuclides of known types and nuclides of unknown types;

[0050] The nuclide attribute acquisition module is used to acquire the semantic attributes of the nuclide based on the recognition result, wherein if the recognition result is a nuclide of a known type, the semantic attributes of the nuclide of the known type are acquired from a predefined nuclide semantic knowledge base; if the recognition result is a nuclide of an unknown type, the semantic attributes of the nuclide of the unknown type are inferred using a pre-trained zero-shot learning model;

[0051] The decision generation module is used to generate a recommended decision based on the properties of the nuclides corresponding to the real-time multimodal data using a predefined decision support model for reference by the operator.

[0052] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, steps of a method for real-time monitoring of radioactive aerosols are implemented.

[0053] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of a method for real-time monitoring of radioactive aerosols.

[0054] The beneficial effects of the present invention are as follows: by collecting real-time multimodal data of aerosols in the target area and fusing the characteristics of these data, more comprehensive and rich monitoring information can be obtained, which not only improves the comprehensiveness and accuracy of the monitoring data, but also can better adapt to complex and changeable environmental conditions; using a pre-trained recognition model to perform radionuclide identification on the fused multimodal data, rapid and accurate identification of known and unknown types of radionuclides can be achieved.

[0055] For unknown types of nuclides, by introducing a zero-shot learning model, the semantic attributes of the unknown nuclides can be inferred based on the existing knowledge base and learning experience, thereby greatly expanding the recognition range and adaptability of the system; finally, based on the attribute information of the nuclides, the predefined decision support model is used to generate recommended decisions, providing strong technical support for operators. It can not only help operators understand the contamination status of radioactive aerosols in a timely manner, but also guide them to take reasonable countermeasures to minimize the risk of radioactive contamination. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0057] Figure 1A schematic flow chart of a method for real-time monitoring of radioactive aerosols provided by an embodiment of the present invention.

[0058] Figure 2 A schematic diagram of working modules of a real-time monitoring system for radioactive aerosols provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0060] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0061] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0062] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0063] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0064] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0065] Example 1, reference Figure 1 , which is the first embodiment of the present invention, provides a method for real-time monitoring of radioactive aerosols, comprising:

[0066] S1: Collect real-time multimodal data of aerosol monitoring in the target area and obtain the fusion features of real-time multimodal data.

[0067] Furthermore, the real-time multimodal data includes real-time collection of multimodal data of radiation data, meteorological data, air quality data and location data through sensor modules deployed in the target area, and the use of feature extraction methods to process each modal data separately. After extracting the features, all single modal features are integrated to form highly generalized fusion features.

[0068] It should be noted that the acquisition of fusion features of real-time multimodal data includes applying a specific feature extraction method to each modality data, specifically:

[0069] Radiation detectors are used to monitor levels of radioactive particles in the air, such as gamma ray count rates.

[0070] Weather stations are used to collect meteorological parameters such as temperature, humidity, wind speed, and wind direction to help understand aerosol diffusion conditions.

[0071] Air quality monitors are used to measure the concentration of particulate matter such as PM2.5 and PM10, as well as gaseous pollutants such as SO2 and NOx, to assist in determining the source and properties of aerosols.

[0072] GPS positioning equipment is used to record the precise geographic coordinates of each monitoring point to facilitate spatial analysis and map visualization.

[0073] At the same time, multiple sensor modules run continuously to collect data at fixed time intervals to achieve real-time monitoring of the environmental conditions in the target area, provide comprehensive and accurate basic data for subsequent data analysis, and enhance the system's ability to respond to emergencies;

[0074] A multimodal fusion network is used to fuse the feature information of each modality, integrate the feature information of each modality, and form a fusion feature that comprehensively reflects the environmental status of the target area. Through the multimodal fusion network, the feature vectors of different modalities are used as input. MMFN contains multiple branches, each branch processes a corresponding modality feature, extracts high-level abstract features, and the outputs of each branch are converged at the fusion layer;

[0075] The following formula is used to extract features from the kth mode:

[0076]

[0077] Among them, f k (x k ) represents the eigenvector of the kth mode, x k Represents the original data of the kth mode, N k represents the number of data points of the kth mode, ρ k represents the weight factor of the kth mode, x ki represents the i-th data point in the k-th mode, μ k represents the center of the kth modal data, φ k represents the nonlinear feature extraction function, σ k represents the standard deviation of the kth modal data;

[0078] In the fusion layer, the outputs of each branch are aggregated. The following is the fusion formula:

[0079]

[0080] Where K represents the total number of modes, W k represents the weight of the kth mode, and H represents the fused feature vector;

[0081] A unified fusion feature vector is generated by weighted summation, and finally the fusion feature vector undergoes further nonlinear transformation and dimensionality reduction to form a highly concentrated and easy-to-interpret fusion feature.

[0082] S2: Use the pre-trained recognition model to perform nuclide recognition on the fusion features of real-time multimodal data to obtain the recognition result.

[0083] Furthermore, the pre-trained recognition model includes analyzing the fusion features through the pre-trained recognition model, outputting the prediction probability of each nuclide type, that is, the confidence, to provide a basis for subsequent nuclide type judgment, and sending the obtained fusion features as input to the trained neural network model. The neural network model is trained using a back propagation algorithm, and the prediction error is minimized by optimizing the loss function to learn the characterization patterns of different nuclide types;

[0084] The forward propagation process includes the input layer receiving the fused feature vector H and introducing the attention mechanism layer to dynamically weight the features of each modality:

[0085] A=Attention(H)

[0086] Among them, A represents the output of the attention mechanism layer, and Attention is the attention function;

[0087] Assume that there are L hidden layers, each with n l Each layer performs feature transformation and nonlinear activation functions to gradually extract higher-level features:

[0088] a (l) =Re(W (l) a (l-1) +b (l) +Dropout(a (l-1) , p l )+A (l-1) )

[0089] Among them, a (l) represents the activation output of the lth layer, Re represents the nonlinear activation function, W (l) represents the weight matrix connecting the l-1th layer and the lth layer, A (l-1) represents the output of the l-1 attention mechanism layer, p l Indicates the Dropout ratio of the lth layer, Dropout(a (l-1) , p l ) represents the random dropout function, b (l) Represents the bias vector of the lth layer; during the training process, the Dropout ratio p l Gradually decrease in each iteration to achieve dynamic adjustment:

[0090] p l =p start -iteration×decay

[0091] Among them, p start Represents the initial Dropout ratio, iteration represents the current number of iterations, and decay represents the decay rate of the Dropout ratio;

[0092] The output layer uses the Softmax function to output the predicted probability of each nuclide type, that is, the confidence

[0093]

[0094] Among them, Softmax means that the output is a function of probability distribution, W (L+1)represents the weight matrix connecting the last hidden layer and the output layer, a (L) represents the activation output of the output layer, b (L+1) represents the bias vector of the output layer, A (L) Represents the output of the current attention mechanism layer;

[0095] The training dataset contains a large amount of labeled multimodal data, each of which is annotated with the actual nuclide type, for model training and verification.

[0096] It should be noted that the fuzzy logic similarity threshold is determined to be 0.8 through a comparison algorithm based on fuzzy logic. When the determined nuclide type matches the confidence of a nuclide type in the known nuclide list, but there is a certain degree of ambiguity, fuzzy logic is used for auxiliary judgment. If the similarity exceeds the threshold, the match is considered successful.

[0097] For the nuclide type similarity evaluation, the similarity threshold is set to 0.7. When the determined nuclide type matches the confidence of a nuclide type in the known nuclide list, but there is a certain degree of difference, the similarity between the two is evaluated. If the similarity exceeds the threshold, the match is considered successful.

[0098] The confidence threshold of the new nuclide type is set to 0.9, and the list of known nuclides is checked regularly. If the confidence of the new nuclide type exceeds the threshold, it is added to the list and the system is updated. According to the above rules, it is judged whether the nuclide type of the real-time multimodal data is known. Different subsequent processing strategies are adopted according to the known / unknown classification of the nuclide type.

[0099] Furthermore, the nuclide identification includes screening out the confidence exceeding a preset confidence threshold from the confidence of each nuclide type corresponding to the real-time multimodal data, finding the nuclide type with the highest confidence in the screened prediction results, treating the nuclide type as the nuclide type identified by this monitoring, judging whether the nuclide type of the real-time multimodal data is known based on the known nuclide list, and obtaining nuclides of known types and nuclides of unknown types;

[0100] The initial confidence threshold is 0.9. Within 5 consecutive iterations, when the number of nuclide types exceeding the confidence threshold exceeds 50%, the confidence threshold increases by 0.05; when it is less than 30%, the confidence threshold decreases by 0.05, but the minimum value is 0.7. After each iteration, the confidence threshold is automatically adjusted according to the above rules.

[0101] The first-level nuclide type confidence threshold is 0.95, the second-level nuclide type confidence threshold is 0.9, the third-level nuclide type confidence threshold is 0.85, and the fourth-level nuclide type confidence threshold is 0.8. Classification is performed according to the nuclide type confidence corresponding to the real-time multimodal data. In the time series of multimodal data, the time series threshold is 3 iteration cycles. Within 3 consecutive iteration cycles, when the confidence of a certain nuclide type continues to increase and exceeds the time series threshold, this nuclide type is given priority.

[0102] Directly obtain the semantic properties of known nuclides or infer the properties of unknown nuclides using zero-shot learning models

[0103] Get a list of known nuclides, which contains all known nuclide types and their related attributes. Compare the determined nuclide type with the known nuclide list. If a match is successful, it is identified as a nuclide of known type; if no match is found, it is considered to be an unknown type of nuclide. Through the known / unknown classification of nuclide types, different processing strategies are adopted according to different types of nuclides.

[0104] S3: Based on the recognition results, the semantic attributes of the nuclides are determined to obtain the semantic attributes of the known types of nuclides, and the semantic attributes of the unknown types of nuclides are inferred using the zero-shot learning model.

[0105] Furthermore, the determining of the semantic attributes of the nuclides includes integrating the sample data and the fusion features corresponding to the nuclides of unknown types, and inputting the fusion features into a pre-trained zero-shot learning model;

[0106] Extract key attributes from the collected sample data to form an attribute matrix, merge the feature vector of the fusion feature corresponding to the unknown type of nuclide with the attribute matrix to form a unified input vector.

[0107] The input vector is sent to the pre-trained zero-shot learning model. The model infers the semantic attributes of the unknown type of nuclides by learning the attribute associations and patterns between different nuclides. The original zero-shot learning model uses the cross-nuclide attribute associations and patterns learned in the training phase to predict the semantic attributes of the unknown type of nuclides. The inference formula is:

[0108] f(x,θ)=S

[0109] Among them, f is the attribute prediction function, θ is the parameter of the zero-shot learning model, S is the semantic attribute of the predicted unknown type of nuclide, and X is the fusion feature vector of the unknown type of nuclide. The original zero-shot learning model is optimized to obtain the improved model, which is expressed as:

[0110]

[0111] in, represents the semantic attribute of the predicted unknown type of nuclide, N represents the number of fused features of the unknown type of nuclide, M represents the number of sample data sets of known nuclides, and x nm represents the fused feature vector of the nth unknown nuclide, θ nm represents the attention weight between the nth unknown nuclide and the mth known nuclide, which is a parameter of the zero-shot learning model. MLP means that a nm Represents the matching score between the fused feature vector of the nth unknown nuclide and the attribute matrix of the mth known nuclide sample data;

[0112] Since the value range of the output S is [0, 1], 0 indicates that it does not have the current semantic attribute, and 1 indicates that it has the current semantic attribute, the possibility of predicting the semantic attribute of the unknown nuclide increases with the increase of the output value.

[0113] It should be noted that the known nuclide sample data set is:

[0114]

[0115] Among them, X s is a known nuclide sample data set, X, Y, Z are the number of known nuclides, and μ is the number of attributes of each nuclide;

[0116] The unknown nuclide fusion feature data set is:

[0117]

[0118] Where X U is the unknown nuclide fusion feature data set, K is the number of fusion features of the unknown nuclides in the matrix,

[0119] The input vector is represented as:

[0120]

[0121] Among them, X' represents the input vector, which contains the known nuclide sample data set and the unknown nuclide fusion feature data set.

[0122] S4: Generate recommended decisions based on the properties of the nuclides corresponding to the real-time multimodal data using a predefined decision support model.

[0123] Furthermore, the generating of the suggested decision includes integrating the basic attributes and extended attributes of the nuclides obtained based on the real-time multimodal data into a comprehensive attribute list; collecting the basic attributes of the nuclides parsed from the real-time multimodal data, integrating the extended attributes, and forming a comprehensive attribute list according to the attribute information in a predefined format and structure;

[0124] Based on the generated comprehensive attribute list, the risk index is calculated to reflect the potential danger level of the nuclide. According to the preset risk level classification standard, the calculated risk index is mapped to a specific risk level, including low risk, medium risk, and high risk;

[0125] RiskIndex(N)=α1·R+α3·T+α4·T x

[0126] Among them, RiskIndex(N) represents the toxicity of the nuclide, α1, α2, α3, and α4 represent the weight coefficients in the risk index calculation, R represents the radioactivity intensity of the nuclide, T represents the half-life of the nuclide, and T x Indicates the toxicity of the nuclide;

[0127] Set the threshold of level 1 risk as RiskIndex l , the threshold of the second-level risk is RiskIndex m , the threshold of the third-level risk is RiskIndex h , when RiskIndex(N)<RiskIndex l When RiskIndex l ≤RiskIndex(N)<RiskIndex m When RiskIndex(N)≥RiskIndex h When the nuclides are classified into three levels of risk, the secondary and tertiary nuclide types will compete with each other in the three-level classification. Through further analysis and comparison, fuzzy logic and similarity evaluation are used to distinguish the nuclide types.

[0128] Establish a policy library containing response strategies. Each strategy corresponds to different risk levels and scenarios. When the risk level is determined, query the policy library for the matching response strategy. According to the actual situation and resource availability, formulate an execution plan for the response strategy, including the execution sequence, participants, and required materials. Execute the selected response strategy, monitor the execution effect, and adjust the strategy or take additional measures when necessary.

[0129] The risk levels and selected response strategies are collated into a report and communicated to operations personnel.

[0130] Embodiment 2, the second embodiment of the present invention, is different from the previous embodiment in that:

[0131] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0132] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0133] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0134] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0135] Example 4, reference Figure 2 , which is an embodiment of the present invention, provides a radioactive aerosol real-time monitoring system, characterized by: comprising a fusion feature acquisition module, a nuclide identification module, a nuclide attribute acquisition module, and a decision generation module;

[0136] A fusion feature acquisition module is used to collect real-time multimodal data of aerosol monitoring in a target area and obtain fusion features of the real-time multimodal data;

[0137] A nuclide identification module, used to perform nuclide identification on the fusion features of the real-time multimodal data using a pre-trained identification model to obtain an identification result, wherein the identification result includes nuclides of known types and nuclides of unknown types;

[0138] a nuclide attribute acquisition module, configured to acquire the semantic attributes of the nuclide based on the recognition result, wherein if the recognition result is a nuclide of known type, the semantic attributes of the nuclide of known type are acquired from a predefined nuclide semantic knowledge base; if the recognition result is a nuclide of unknown type, the semantic attributes of the nuclide of unknown type are inferred using a pre-trained zero-shot learning model;

[0139] The decision generation module is used to generate a suggested decision based on the properties of the nuclides corresponding to the real-time multimodal data using a predefined decision support model for reference by operators.

[0140] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for real-time monitoring of radioactive aerosols, characterized in that: include, Collect real-time multimodal data of aerosol monitoring in the target area and obtain the fusion features of real-time multimodal data; Use the pre-trained recognition model to perform nuclide recognition on the fusion features of real-time multimodal data to obtain the recognition result; Based on the recognition results, the semantic attributes of the nuclides of known types are obtained by judging the semantic attributes of the nuclides, and the semantic attributes of the nuclides of unknown types are inferred using the zero-shot learning model; Based on the properties of the nuclides corresponding to the real-time multimodal data, a predefined decision support model is used to generate recommended decisions.

2. A method for real-time monitoring of radioactive aerosols according to claim 1, characterized in that: The real-time multimodal data includes multimodal data of radiation data, meteorological data, air quality data and location data collected in real time by sensor modules deployed in the target area, and feature extraction methods are used to process each modal data separately. After extracting the features, all single modal features are integrated to form highly generalized fusion features.

3. A method for real-time monitoring of radioactive aerosols according to claim 2, characterized in that: The method of obtaining fusion features of real-time multimodal data includes applying a specific feature extraction method to each modality data, specifically: The radiation data uses spectrum analysis and statistical methods to extract peak energy and energy spectrum distribution characteristics; Meteorological data by calculating trends, cyclical changes and outliers; Air quality data identifies concentration fluctuation patterns by analyzing the time series characteristics of pollutants; The location data were analyzed using geographic information system by calculating the distance matrix and spatial clustering between each monitoring point; A multimodal fusion network is used to fuse the feature information of each modality, integrate the feature information of each modality, and form a fusion feature that comprehensively reflects the environmental status of the target area. Through the multimodal fusion network, the feature vectors of different modalities are used as input. MMFN contains multiple branches, each branch processes a corresponding modality feature, extracts high-level abstract features, and the outputs of each branch are converged at the fusion layer; In the fusion layer, the outputs of each branch are aggregated and a unified fusion feature vector is generated through weighted summation. Finally, the fusion feature vector undergoes further nonlinear transformation and dimensionality reduction to form a highly concentrated and easy-to-interpret fusion feature.

4. A method for real-time monitoring of radioactive aerosols according to claim 3, characterized in that: The pre-trained recognition model includes: analyzing the fusion features through the pre-trained recognition model, outputting the prediction probability of each nuclide type, that is, the confidence, to provide a basis for subsequent nuclide type judgment, and sending the obtained fusion features as input to the trained neural network model. The neural network model is trained using a back propagation algorithm, and the prediction error is minimized by optimizing the loss function to learn the characterization patterns of different nuclide types; The forward propagation process includes the input layer receiving the fused feature vector H and introducing the attention mechanism layer to dynamically weight the features of each modality: A=Attention(H) Among them, A represents the output of the attention mechanism layer, and Attention is the attention function; Assume that there are L hidden layers, each with n l Each layer performs feature transformation and nonlinear activation functions to gradually extract higher-level features: a (l) =Re(W (l) a (l-1) +b (l) +Dropout(a (l-1) ,p l )+A (l-1) ) Among them, a (l) represents the activation output of the lth layer, Re represents the nonlinear activation function, W (l) represents the weight matrix connecting the l-1th layer and the lth layer, A (l-1) represents the output of the l-1 attention mechanism layer, p l Indicates the Dropout ratio of the lth layer, Dropout(a (l-1) ,p l ) represents the random dropout function, b (l) Represents the bias vector of the lth layer; during the training process, the Dropout ratio p l Gradually decrease in each iteration to achieve dynamic adjustment: p l =p start -iteration×decay Among them, p start Represents the initial Dropout ratio, iteration represents the current number of iterations, and decay represents the decay rate of the Dropout ratio; The output layer uses the Softmax function to output the predicted probability of each nuclide type, that is, the confidence Among them, Softmax means that the output is a function of probability distribution, W( L+1 ) represents the weight matrix connecting the last hidden layer and the output layer, a( L ) represents the activation output of the output layer, b( L+1 ) represents the bias vector of the output layer, A( L ) represents the output of the current attention mechanism layer; The training dataset contains a large amount of labeled multimodal data, each of which is annotated with the actual nuclide type, for model training and verification.

5. A method for real-time monitoring of radioactive aerosols according to claim 4, characterized in that: The performing of nuclide identification includes screening out the confidence exceeding a preset confidence threshold from the confidence of each nuclide type corresponding to the real-time multimodal data, finding the nuclide type with the highest confidence in the screened prediction results, treating the nuclide type as the nuclide type identified by this monitoring, judging whether the nuclide type of the real-time multimodal data is known based on the known nuclide list, and obtaining nuclides of known types and nuclides of unknown types; The initial confidence threshold is 0.

9. Within 5 consecutive iterations, when the number of nuclide types exceeding the confidence threshold exceeds 50%, the confidence threshold increases by 0.05; when it is less than 30%, the confidence threshold decreases by 0.05, but the minimum value is 0.

7. After each iteration, the confidence threshold is automatically adjusted according to the above rules. The first-level nuclide type confidence threshold is 0.95, the second-level nuclide type confidence threshold is 0.9, the third-level nuclide type confidence threshold is 0.85, and the fourth-level nuclide type confidence threshold is 0.

8. Classification is performed according to the nuclide type confidence corresponding to the real-time multimodal data. In the time series of multimodal data, the time series threshold is 3 iteration cycles. Within 3 consecutive iteration cycles, when the confidence of a certain nuclide type continues to increase and exceeds the time series threshold, this nuclide type is given priority. Directly obtain the semantic attributes of known nuclides or use the zero-shot learning model to infer the attributes of unknown nuclides to obtain a list of known nuclides, which contains all known nuclide types and their related attributes. Compare the determined nuclide type with the known nuclide list. If a match is successful, it is identified as a nuclide of a known type; if no match is found, it is considered to be an unknown type of nuclide. Through the known / unknown classification of nuclide types, different processing strategies are adopted according to different types of nuclides.

6. A method for real-time monitoring of radioactive aerosols according to claim 5, characterized in that: Determining the semantic attributes of the nuclide includes integrating the sample data and the fusion features corresponding to the nuclide of unknown type, and inputting the fusion features into a pre-trained zero-shot learning model; Extract key attributes from the collected sample data to form an attribute matrix, merge the feature vector of the fusion feature corresponding to the unknown type of nuclide with the attribute matrix to form a unified input vector. The input vector is sent to the pre-trained zero-shot learning model. The model infers the semantic attributes of the unknown type of nuclides by learning the attribute associations and patterns between different nuclides. The original zero-shot learning model uses the cross-nuclide attribute associations and patterns learned in the training phase to predict the semantic attributes of the unknown type of nuclides. The inference formula is: f(X,θ)=S Among them, f is the attribute prediction function, θ is the parameter of the zero-shot learning model, S is the semantic attribute of the predicted unknown type of nuclide, and X is the fused feature vector of the unknown type of nuclide; the original zero-shot learning model is optimized to obtain the improved model; Since the value range of the output S is [0,1], 0 means that it does not have the current semantic attribute, and 1 means that it has the current semantic attribute. The possibility of predicting the semantic attribute of the unknown nuclide increases with the increase of the output value.

7. A method for real-time monitoring of radioactive aerosols according to claim 6, characterized in that: The generating of the recommended decision includes integrating the basic attributes and extended attributes of the nuclides obtained based on the real-time multimodal data into a comprehensive attribute list; Collect the basic properties of nuclides parsed from real-time multimodal data, integrate the extended properties, and form a comprehensive property list according to the predefined format and structure; Based on the generated comprehensive attribute list, the risk index is calculated to reflect the potential danger level of the nuclide. According to the preset risk level classification standard, the calculated risk index is mapped to a specific risk level, including low risk, medium risk, and high risk; Establish a policy library containing response strategies. Each strategy corresponds to different risk levels and scenarios. When the risk level is determined, query the policy library for the matching response strategy. According to the actual situation and resource availability, formulate an execution plan for the response strategy, including the execution sequence, participants, and required materials. Execute the selected response strategy, monitor the execution effect, and adjust the strategy or take additional measures when necessary. The risk levels and selected response strategies are collated into a report and communicated to operations personnel.

8. A system using a real-time monitoring method for radioactive aerosols as claimed in any one of claims 1 to 7, characterized in that: It includes fusion feature acquisition module, nuclide identification module, nuclide attribute acquisition module and decision generation module; The fusion feature acquisition module is used to collect real-time multimodal data of aerosol monitoring in the target area and obtain fusion features of the real-time multimodal data; The nuclide identification module is used to perform nuclide identification on the fusion features of the real-time multimodal data using a pre-trained identification model to obtain an identification result, wherein the identification result includes nuclides of known types and nuclides of unknown types; The nuclide attribute acquisition module is used to acquire the semantic attributes of the nuclide based on the recognition result, wherein if the recognition result is a nuclide of a known type, the semantic attributes of the nuclide of the known type are acquired from a predefined nuclide semantic knowledge base; if the recognition result is a nuclide of an unknown type, the semantic attributes of the nuclide of the unknown type are inferred using a pre-trained zero-shot learning model; The decision generation module is used to generate a suggested decision based on the properties of the nuclides corresponding to the real-time multimodal data using a predefined decision support model for reference by operators.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a method for real-time monitoring of radioactive aerosols according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a real-time monitoring method for radioactive aerosols according to any one of claims 1 to 7 are implemented.

Images